KernelRun analyzes your AWS, GCP, and Azure resource usage and generates right-sizing schedules automatically. Engineering teams that connect in the first billing cycle typically see 28–41% reduction in compute spend.
KernelRun connects to your cloud accounts via read-only IAM role, ingests 90 days of CloudWatch metrics, and produces concrete scheduling actions — not recommendations you have to interpret.
Scans EC2, RDS, ECS, and GKE workloads across all regions. Identifies idle instances, orphaned snapshots, and over-provisioned reserved capacity within 15 minutes of connecting.
Generates on/off schedules based on actual usage patterns — not business hours defaults. Schedules are proposed, not applied, until your team approves them via Slack or the web console.
Maps costs to git repositories, Kubernetes namespaces, and Jira projects using tag inference and commit-level telemetry. Finance gets line items; engineers get context.
Flags cost spikes within 4 hours of first observation using a multi-variate baseline model trained on your account's own 90-day history. No static thresholds to maintain.
Compares p95 CPU and memory utilization against 47 EC2 instance types and recommends the smallest type that satisfies observed peak demand with a configurable headroom buffer.
All connections use read-only IAM roles with least-privilege policies. No credentials stored. Every schedule change requires approval. Rollback available within 24 hours.
Create a read-only IAM role using our CloudFormation template. KernelRun gets access to CloudWatch metrics and Cost Explorer data — nothing else.
The scheduler ingests 90 days of CloudWatch utilization data per resource and builds a usage profile per service, environment, and team. Takes 10–20 minutes depending on account size.
KernelRun generates a schedule proposal: which resources to stop, when, and for how long. Each proposal shows the projected monthly savings and the utilization evidence behind it.
Your team approves, edits, or declines each proposal via Slack or the dashboard. Approved schedules run on AWS Lambda. KernelRun monitors for exceptions and alerts within 4 hours.
Most cloud cost tools show you which services are expensive. KernelRun maps that cost to specific git repositories, Kubernetes deployments, and pull requests — so the engineering team responsible gets the data, not just the finance team.
Attribution uses tag inference when native tags are missing, and cross-references AWS Cost Explorer with GitHub commit history to fill gaps in tagging coverage.
Explore the Platform
Dev, staging, and QA environments run 24/7 but are used roughly 9 hours a day. KernelRun identifies these environments from tagging patterns and proposes off-hours shutdown schedules. Average saving: $1,200/month per environment.
Production instances provisioned during peak load often run at 12–18% CPU the rest of the time. KernelRun identifies the smallest instance type that covers observed p95 load plus your configured headroom percentage.
Identifies read replicas with zero traffic, multi-AZ configurations on non-critical databases, and ElastiCache clusters whose hit rate falls below a configurable threshold — three categories that regularly go unreviewed.
Connect your first AWS account in 4 minutes. Analysis results appear in under 20 minutes.
Request a Demo